def test_package_common_3(self):
maintain_config = common.load_yaml_file("test-maintain.yaml")
cluster_object_model = common.load_yaml_file(
"test-generated-cluster-object-model.yaml")
node_config = cluster_object_model['machine']['machine-list'][
'worker-01']
common.maintain_package_wrapper(cluster_object_model, maintain_config,
node_config, "unittest-common-3")
self.assertTrue(
os.path.exists("parcel-center/1.2.3.2/unittest-common-3.tar"))
package = tarfile.open("parcel-center/1.2.3.2/unittest-common-3.tar",
"r:")
package.extractall()
self.assertTrue(os.path.exists("unittest-common-3/"))
target_file_list = ["testfile1.sh"]
package_file_list = os.listdir("unittest-common-3/")
self.assertListEqual(sorted(target_file_list),
sorted(package_file_list))
shutil.rmtree("unittest-common-3/")
common.maintain_package_cleaner(node_config)
self.assertFalse(os.path.exists("parcel-center/1.2.3.2"))
self.assertTrue(os.path.exists("parcel-center"))
shutil.rmtree("parcel-center/")
def test_etcdfix_conf_validation_node_config_validation(self):
node_list = common.load_yaml_file(
"data/data_maintainlib_etcdfix/test_node_list_config.yaml")
cluster_config = common.load_yaml_file(
"data/data_maintainlib_etcdfix/generated-cluster-object-model-ok.yaml"
)
node_config = node_list['machinelist']['ok-machine-node']
validation = etcdfix.etcdfix_conf_validation(cluster_config,
node_config)
self.assertTrue(validation.node_conf_validation())
node_config = node_list['machinelist']['miss-node-name']
validation = etcdfix.etcdfix_conf_validation(cluster_config,
node_config)
self.assertFalse(validation.node_conf_validation())
node_config = node_list['machinelist']['miss-host-ip']
validation = etcdfix.etcdfix_conf_validation(cluster_config,
node_config)
self.assertFalse(validation.node_conf_validation())
node_config = node_list['machinelist']['wrong-host-ip']
validation = etcdfix.etcdfix_conf_validation(cluster_config,
node_config)
self.assertFalse(validation.node_conf_validation())
node_config = node_list['machinelist']['wrong-ssh-port']
validation = etcdfix.etcdfix_conf_validation(cluster_config,
node_config)
self.assertFalse(validation.node_conf_validation())
node_config = node_list['machinelist']['miss-user-name']
validation = etcdfix.etcdfix_conf_validation(cluster_config,
node_config)
self.assertFalse(validation.node_conf_validation())
node_config = node_list['machinelist']['miss-password']
validation = etcdfix.etcdfix_conf_validation(cluster_config,
node_config)
self.assertFalse(validation.node_conf_validation())
node_config = node_list['machinelist']['miss-etcd-id']
validation = etcdfix.etcdfix_conf_validation(cluster_config,
node_config)
self.assertFalse(validation.node_conf_validation())
def setup_logging():
"""
Setup logging configuration.
"""
configuration_path = "sysconf/logging.yaml"
logging_configuration = pai_common.load_yaml_file(configuration_path)
logging.config.dictConfig(logging_configuration)
def test_etcdfix_conf_validation_cluster_config_validation(self):
node_list = common.load_yaml_file(
"data/data_maintainlib_etcdfix/test_node_list_config.yaml")
node_config = node_list['machinelist']['ok-machine-node']
cluster_config = common.load_yaml_file(
"data/data_maintainlib_etcdfix/generated-cluster-object-model-ok.yaml"
)
validation = etcdfix.etcdfix_conf_validation(cluster_config,
node_config)
self.assertTrue(validation.cluster_conf_validation())
cluster_config = common.load_yaml_file(
"data/data_maintainlib_etcdfix/generated-cluster-object-model-miss-master.yaml"
)
validation = etcdfix.etcdfix_conf_validation(cluster_config,
node_config)
self.assertFalse(validation.cluster_conf_validation())
cluster_config = common.load_yaml_file(
"data/data_maintainlib_etcdfix/generated-cluster-object-model-miss-node-config.yaml"
)
validation = etcdfix.etcdfix_conf_validation(cluster_config,
node_config)
self.assertFalse(validation.cluster_conf_validation())
cluster_config = common.load_yaml_file(
"data/data_maintainlib_etcdfix/generated-cluster-object-model-wrong-node-config.yaml"
)
validation = etcdfix.etcdfix_conf_validation(cluster_config,
node_config)
self.assertFalse(validation.cluster_conf_validation())
cluster_config = common.load_yaml_file(
"data/data_maintainlib_etcdfix/generated-cluster-object-model-inconsistent-node-config.yaml"
)
validation = etcdfix.etcdfix_conf_validation(cluster_config,
node_config)
self.assertFalse(validation.cluster_conf_validation())
def test_etcdfix_conf_validation_cluster_config_validation(self):
node_list = common.load_yaml_file(
"data/data_maintainlib_etcdfix/test_node_list_config.yaml")
node_config = node_list['machinelist']['ok-machine-node']
cluster_config = common.load_yaml_file(
"data/data_maintainlib_etcdfix/test_cluster_config_ok.yaml")
validation = etcdfix.etcdfix_conf_validation(cluster_config,
node_config)
self.assertTrue(validation.cluster_conf_validation())
cluster_config = common.load_yaml_file(
"data/data_maintainlib_etcdfix/test_cluster_config_miss_master_list.yaml"
)
validation = etcdfix.etcdfix_conf_validation(cluster_config,
node_config)
self.assertFalse(validation.cluster_conf_validation())
cluster_config = common.load_yaml_file(
"data/data_maintainlib_etcdfix/test_cluster_config_miss_node_config.yaml"
)
validation = etcdfix.etcdfix_conf_validation(cluster_config,
node_config)
self.assertFalse(validation.cluster_conf_validation())
cluster_config = common.load_yaml_file(
"data/data_maintainlib_etcdfix/test_cluster_config_wrong_node_config.yaml"
)
validation = etcdfix.etcdfix_conf_validation(cluster_config,
node_config)
self.assertFalse(validation.cluster_conf_validation())
cluster_config = common.load_yaml_file(
"data/data_maintainlib_etcdfix/test_cluster_config_inconsistent_node_config.yaml"
)
validation = etcdfix.etcdfix_conf_validation(cluster_config,
node_config)
self.assertFalse(validation.cluster_conf_validation())
def load_cluster_config(config_path):
return pai_common.load_yaml_file(config_path)
def main():
parser = argparse.ArgumentParser()
parser.add_argument('-p',
'--path',
required=True,
help='path of cluster configuration file')
parser.add_argument('-a',
'--action',
required=True,
default=None,
help="action to maintain the cluster")
parser.add_argument('-f',
'--file',
default=None,
help="An yamlfile with the nodelist to maintain")
args = parser.parse_args()
logger.info("Begin option validation! ")
if option_validation(args) == False:
return
logger.info("Pass option validation! ")
config_path = args.path
cluster_config = logClusterObjectModel(config_path)
master_list = cluster_config['mastermachinelist']
etcd_cluster_ips_peer, etcd_cluster_ips_server = generate_etcd_ip_list(
master_list)
# ETCD will communicate with each other through this address.
cluster_config['clusterinfo'][
'etcd_cluster_ips_peer'] = etcd_cluster_ips_peer
# Other service will write and read data through this address.
cluster_config['clusterinfo'][
'etcd_cluster_ips_server'] = etcd_cluster_ips_server
cluster_config['clusterinfo']['etcd-initial-cluster-state'] = 'new'
if args.action == 'add':
logger.info("Begin to add new nodes to PAI cluster.")
#Todo in the future we should finish the following two line
#cluster_config = get_cluster_configuration()
#node_list_config = get_node_list_config()
node_list_config = pai_common.load_yaml_file(args.file)
maintain_nodes(cluster_config, node_list_config, args.action)
#up_data_cluster_configuration()
logger.info("New nodes have been added.")
return
if args.action == 'remove':
logger.info("Begin to remove nodes from PAI cluster.")
# Todo in the future we should finish the following two line
# cluster_config = get_cluster_configuration()
# node_list_config = get_node_list()
node_list_config = pai_common.load_yaml_file(args.file)
maintain_nodes(cluster_config, node_list_config, args.action)
# up_data_cluster_configuration()
logger.info("Nodes have been removed.")
return
if args.action == 'repair':
logger.info("Begin to repair the target nodes.")
# Todo in the future we should finish the following two line
# cluster_config = get_cluster_configuration()
# node_list_config = get_node_list()
node_list_config = pai_common.load_yaml_file(args.file)
maintain_nodes(cluster_config, node_list_config, args.action)
logger.info("The nodes have been repaired.")
return
if args.action == 'clean':
logger.info("Begin to clean up whole cluster.")
maintain_cluster(cluster_config, job_name=args.action, clean=True)
logger.info("Clean up job finished")
return
if args.action == 'etcdfix':
logger.info("Begin to fix broken etcd server.")
node_list_config = pai_common.load_yaml_file(args.file)
logger.debug("FIX ETCD on {0}".format(str(node_list_config)))
if len(node_list_config['machinelist']) != 1:
logger.error(
"etcdfix can't fix more than one machine everytime. Please fix them one by one!"
)
sys.exit(1)
maintain_nodes(cluster_config, node_list_config, args.action)
logger.info("Etcd has been fixed.")
return
if args.action == 'deploy':
logger.info("Begin to initialize PAI.")
maintain_cluster(cluster_config, job_name=args.action, clean=True)
logger.info("Finish initializing PAI.")
if args.action == 'install_kubectl':
logger.info("Begin to install kubectl.")
maintain_cluster(cluster_config, job_name="kubectl_install")
logger.info("Kubectl has been installed.")
logger.info("Maintenance Finished!")
def main():
setup_logging()
parser = argparse.ArgumentParser()
parser.add_argument('-p', '--path', required=True, help="cluster configuration's path")
parser.add_argument('-c', '--clean', action="store_true", help="clean the generated script")
parser.add_argument('-d', '--deploy', action="store_true", help="deploy all the service")
parser.add_argument('-s', '--service', default='all', help="bootStrap a target service")
parser.add_argument('-a', '--action', choices=['add', 'remove'], help="action to maintain services")
parser.add_argument('-f', '--file', help="An yamlfile with the nodelist to maintain services on nodes")
args = parser.parse_args()
# step 1: load configuration from yaml file.
config_path = args.path
cluster_config = loadClusterObjectModel(config_path)
service_config = load_yaml_config("service.yaml")
# step 2: generate base64code for secret.yaml and get the config.json of docker after logining
generate_secret_base64code(cluster_config[ "clusterinfo" ][ "dockerregistryinfo" ])
generate_docker_credential(cluster_config[ "clusterinfo" ][ "dockerregistryinfo" ])
# step 3: generate image url prefix for yaml file.
generate_image_url_prefix(cluster_config[ "clusterinfo" ][ "dockerregistryinfo" ])
if 'docker_tag' not in cluster_config['clusterinfo']['dockerregistryinfo']:
cluster_config['clusterinfo']['dockerregistryinfo']['docker_tag'] = 'latest'
# step 4: generate configuration of hadoop queues
generate_configuration_of_hadoop_queues(cluster_config)
# step 5: generate templatefile
if args.action == 'add' or args.action == 'remove':
node_list_config = pai_common.load_yaml_file(args.file)
cluster_config, service_config = generate_new_nodes_template_file(
cluster_config, service_config, node_list_config)
elif args.service == 'all':
copy_arrangement(service_config)
generate_template_file(cluster_config, service_config)
else:
copy_arrangement_service(args.service, service_config)
generate_template_file_service(args.service, cluster_config, service_config)
# step 6: Bootstrap service.
# Without flag -d, this deploy process will be skipped.
if args.action == 'add':
bootstrap_service_new_nodes(service_config)
elif args.action == 'remove':
remove_service(cluster_config, service_config, node_list_config)
elif args.deploy:
if args.service == 'all':
# dependency_bootstrap will auto-start all service in a correctly order.
bootstrap_service(service_config)
else:
# Single service startup will ignore the dependency. User should ensure the operation is in a correctly order. This option is mainly designed for debug.
single_service_bootstrap(args.service, service_config)
# Optional : clean all the generated file.
if args.clean:
clean_up_generated_file(service_config)
